An Analysis of the Wage Impact of Trade Unions in the UK Public and Private Sectors.

Transcription

1 An Analysis of the Wage Impact of Trade Unions in the UK Public and Private Sectors. David G. Blanchflower Bruce V. Rauner '78 Professor of Economics Dartmouth College and NBER Website: March 23 rd 2005 I thank Alex Bryson, Richard Freeman, Steve Machin, David Metcalf, John Pencavel and Andrew Oswald for helpful comments

2 This paper examines changes in unionization and the impact unions have had on wages over the last decade or so in the United Kingdom. The UK has seen a steady decline in union density since the end of the 1970s with broadly similar percentage point declines occurring in both the public and private sectors. I examine the characteristics of a nonunion workforce that exists in the government sector that are less likely to have permanent jobs, have shorter tenure and have lower pay than their public sector union counterparts. A significant proportion of these non-union public sector workers are employed in workplaces where their occupation is covered by a union collective bargaining agreement, but they are still paid less than their union counterparts at that workplace. The non-union public sector workforce is more similar to the non-union private sector workforce than it is to the unionized public sector workforce. In part this appears to arise from the fact that some of these non-union workers who report that they work in the public sector are actually employed by private sector firms. The union/nonunion wage gap in the public sector in the UK is double that existing in the private sector. Women in the UK benefit the most from belonging to a union in terms of higher wages and especially so in the public sector. The low paying jobs are held by public sector workers who are not members of unions. First, I examine changes in union density across OECD countries and identify a number of explanations. Second, I present the facts regarding who joins unions and how that has changed over time in both the UK public and private sectors. I report the finding across all the major OECD countries, that public sector union density rates are much higher than private sector rates. Third, I make use of large individual level micro-data files from the UK Labour Force Surveys of to estimate a series of wage equations and I find that the union wage differential or wage gap is especially high for women. Further, I find a higher union wage effect in the public sector than in the private sector which on its face seems unlikely given the high levels of union recognition and coverage operating in the public sector. I also examine data from the General Household Survey of 1983 which shows an overall decline over the last twenty years in the overall size of the differential driven mostly by a decrease in the size of the private sector differential alongside an increase over time in the size of the public sector wage gap. Fourth, I examine who the non-union workers are in the UK public sector and show that their characteristics are more comparable to non-union workers in the private sector than they are to public sector union workers. Fifth, I use establishment level data from the Workplace Industrial Relations Survey Time Series Dataset, to document the extensive use that public sector workplaces make of outsourcing as well as temporary workers and workers on fixed-term contracts who are often non-union. Sixth, I explore the question of whether this large public sector differential that exists currently is real or simply an artifact of measurement error in the LFS data. In the LFS there is a tendency for individuals working for private sector subcontractors to misreport being in the public sector which results in estimates of the numbers of workers employed in the public sector being nearly 1.2 million higher when the LFS data are used compared with estimates derived from public sector organizations. I find that the differential is real and only in part impacted by these data measurement problems. There is a substantial and significant wage gap between the earnings of members and nonmembers in workplaces covered by 1

3 union bargaining agreements in both the private and public sectors. conclusions. Finally, I draw 1. Background - changes in unionization rates over time Over the last two or three decades, the social and economic environment became increasingly hostile to unionism internationally and to many traditional union practices and policies. In this environment, unions in the UK and the US in particular suffered dramatic losses, comparable to those experienced in the 1920s and 1930s. There is now evidence showing that unions are in decline in most other OECD countries also but there are several Northern European countries - Belgium, Denmark, Iceland, Finland and Sweden - where union density rates have increased over time. Table 1 reports levels of union density for 1970, 1980, 1990 and 2000 for 20 OECD countries as well as estimates of the degree of union coverage in There are some problems in comparing the data because there are differences in how the data are collected: some are taken from administrative sources and some from surveys; some countries have unemployed and retired workers included in their estimates (e.g. Canada) and some do not. 1 Not withstanding all of these caveats it is apparent that the United States overall level of density is atypical of the OECD. Other countries with low densities such as the Netherlands and Spain have had very different trends and histories. Even in countries like the UK that have had considerable declines in density over the last two decades, the percentage of workers that are members is still considerably higher than it is in the United States (e.g. 14% in the U.S. in 2000 compared with 31% in the UK). With the exception of Turkey, which has seen a 5 percentage point increase in density since 1990 and Korea which has seen little change from its very low starting level of 13% in 1970, unions have been in decline in all of the newer member countries of the OECD. Unions have been in decline in Spain, Greece, Portugal and Mexico as well as in all four of the ex-communist country members (Czech Republic, Hungary, Poland and the Slovak Republic). The unweighted OECD average declined from 42% in 1970 to 34% in 2000 and with an unchanging sample of countries the average fell to 36%. In an earlier paper (Blanchflower and Freeman, 1992) countries were classified into four groupings according to whether the country had sharp rises in density; rises in the 70s but stable in the 1980s; rises in the 1970s but declines in the 1980s and declining density. With the availability of data into the twenty first century it is necessary to adapt that taxonomy it is appropriate to now classify the main OECD countries into three groups that showed increases, decreases and little change in density from Several new countries now join the UK and the US in the declining density category - Australia, Germany, Ireland, Luxembourg, New Zealand and Switzerland. The new classification of countries used is thus as follows, based on data reported in Table 1: a) unchanging density -- Canada, Italy, Korea, Norway and Turkey. 1 For a discussion of the problems of comparability of union membership data across countries, see OECD (1991). 2

4 b) sharp rises in density -- Belgium, Denmark, Iceland, Finland and Sweden. c) declining density - Australia, Austria, Czech Republic, France, Germany, Greece, Hungary, Ireland, Japan, Luxembourg, Mexico, the Netherlands, New Zealand, Poland, Portugal, Slovak Republic, Spain, Switzerland, the United Kingdom and the United States. The UK and especially the US experience of rapidly declining unionization looks much more similar to that of its OECD partners now than it did in the early 1990s. In part these declines appear to have been driven by global changes and in part by government action. The countries that have had increases in density with essentially universal coverage are small corporatist countries that have governments that are sympathetic to unions. The group of five countries that have seen approximately unchanging density includes three countries - Canada, Norway and Italy whose governments have generally been union friendly. Flanagan (2005) has recently argued that the hypothesis that declining U.S. unionization results from distinctive U.S. policies and institutions cannot be established by claiming the U.S. membership decline is unique (p. 36). Flanagan goes on to argue that there is strong evidence that the main reason for declining density is declining demand for union representation among nonunion workers around the world. With many of the benefits formerly associated with union contracts now provided by public policy or modern human resource management policies and with global competition often constraining the gains achievable through domestic collective bargaining, many workers appear to doubt the ability of unions to provide what they want from their jobs and prefer less conflictual approaches to collective activities at the workplace (p.60). Pencavel (2003), in a recent examination of the decline of unions in Britain suggested that its cause was the combination of 1) the changed legal environment including the virtual elimination of the closed shop, 2) the abandonment of full-employment macropolicies which meant that organized labor had to operate in a more inhospitable environment 3) the rigors of considerably greater product market competition. He goes on to argue that despite what he calls the surprising retreat of union Britain there does not appear to be a pervasive, unmet demand for union representation (p. 225). Bryson (2003) examines data from the British Worker Representation and Participation Survey of 2001 and confirmed that finding. When non-union members were asked if someone from the union at your workplace asked you to join how likely is it that you would do so?, only 10% said very likely and 26% said quite likely. Charlwood (2002) found very similar responses in his analysis of responses to the same question in the 1998 British Social Attitudes Survey. What is clear is that there has been a decline in private sector density in many OECD countries and as a result public sector density is now markedly higher than private sector 3

5 density. 2 In many OECD countries, including the US and the UK, and several excommunist countries (e.g. Bulgaria, Latvia, Russia and Slovenia) and even a few developing countries (e.g. South Africa, Mexico and Chile), the public sector unionization rate is more than twenty percentage points higher than the rate prevailing in the private sector. Visser (2003) reported union density estimates for a number of countries by public and private sector. A summary of the results from his Table 11.8 are provided below. Private Public Australia (1998) Austria (1998) Belgium (1991) Canada (2000) Finland (1989) France (1993) 4 25 Germany (1997) Israel (1997) Italy (1997) Japan (1995) Netherlands (1997) Norway (1995) Poland (1999) Spain (1997) Sweden (1997) Switzerland (1988) Part of the reason for higher union membership in the public sector may be because governments are seen as exemplary employers who value procedural fairness highly. Moreover, there are lower entry barriers to unions in public sector workplaces because levels of employer hostility to unionization are lower and because public sector workplaces tend to be larger than private sector ones. Another reason is that the government tends to engage in national or sectoral pay bargaining in these sectors, raising the benefits to unionization. However, there is evidence from a number of countries that union membership is in decline in both the public and private sectors. Hence, despite the fixed effect of less hostile employers in the public sector, employees do seem to vote with their feet in both sectors in much the same way. This implies either that a) the net returns to membership have fallen in similar ways in both sectors or b) that unions are facing organizational constraints in the public sector as in the private sector which means they can t reach new entrants to the labor market. Freeman (1988) argued that incentives for employers to oppose unions in the US public sector appear to be lower than in the private sector. First, he argued, because public 2 In the US 1973 public sector density was 23.0% compared with 24.2% in the private sector. In 2004 the rates were 36.4% and 7.9% respectively. Source: Hirsch and Macpherson (2003). The data are downloadable at 4

6 sector workers wield political power there is potential for public sector unions to shift demand outward through the political process. Consistent with this Fiorito et al (1996), in a study of worker attitudes to unions in the US private and public sectors, found that employer resistance and retaliation in response to unionization efforts were much less likely to perceive a coercive response against employees as a result of a union organizing drive (p.474). Second, the cost of illegal opposition, which has tended to occur in the United States, is likely to be greater for public officials. Third, unions can help public sector employers increase budgets through lobbying for additional public expenditures. Fourth, wage premiums, Freeman argued were smaller in the public sector, but as we show below, is not the case now in the UK Who belongs to unions? We now turn to examine UK data on the characteristics of union members in the public and private sectors. As background, the number of economically active people went up from 25.9 million in 1980 to 28.9 million at the end of 2004 (+11.6%). For Great Britain the union density rate fell from 32.2% in 1992 to 26.4% in There was an absolute decline of more than 700,000 in the total number of members (7,857,000 members between 1992 to 7,136,000 and 2003). 4 UK population rose from 56.4 million in 1981 to 59.2 million in 2002 (+5%). Figure 1 shows changes in union density in the UK over a hundred year period from The density rate peaked in 1920 at 38.7% but then fell to a low of 23.0% in 1933 only to rise steadily until 1978, reaching its high watermark of 52.4% in Union density in the UK has been in a downward spiral since that time. The proportion of workers that are union members currently existing at the time of writing in the UK (29.3%) is approximately the same as previously existed in the UK in 1938, 1925 and The density rate prevailing in the UK at the end of the twentieth century was over seventeen percentage points higher than it was at the beginning of the century. Data at the level of the individual on union membership are available in the UK from the Labor Force Survey every year since Union data are only reported in one of the four quarterly sweeps of the survey in the Autumn of each year. Table 2 provides various details of the characteristics of union members in 2003 using these LFS data. Union density is 29.3%: the rates by gender are little different and the membership of blacks is higher than that of whites. This is not true of other racial groups. The young are less likely to be members of unions. Membership rates in manufacturing and construction are lower than average in UK. There is very little union presence in Hotels and restaurants. Public sector unionism is especially high. The more educated have higher unionization rates than the less educated. Membership rates are low for workers in temporary jobs and are very high for workers with long tenure. 3 In Blanchflower and Bryson (2004) we showed that there are now significant union wage gaps in the US in both the public and private sectors of 15% and 17% respectively for the years See Trade union membership, 2003, July 2004 published by Department of Trade and Industry, UK, Table 2. 5

7 Table 3 provides details of changes in overall density rates as well as in the public and private sector by year since 1993 for Great Britain using data from the Labour Force Surveys. All three columns show steady declines. Density in both the private and public sectors is approximately 6 percentage points lower in both sectors in 2003 than it was in Given the lower starting rate in the private sector this means the percentage decline is greater there (-24.6% compared to 8.7% in the public). The second part of Table 3 shows that union density rates are higher in the north than in the south and are especially low in the South East. In every region private density rates are below public. The rankings of regions are broadly similar in the two sectors, with the major exception of Wales which ranks second in the private sector and twelfth in the public. The North East ranks highest on both measures and the South East lowest on both. A recent article published by the Office of National Statistics (Hicks et al, 2005) has produced a set of estimates of public sector employment and has noted there is a discrepancy of more than 1.2 million workers between estimates of public sector employment derived from the Labour Force Survey, based on households, and the numbers derived from administrative and survey data from public sector organizations. 5 In 2004Q1 there were 5,746,000 public sector employees in the UK using data from public sector organizations compared with 6,907,000 using data from the LFS (Hicks et al, 2005, Table 8). The numbers using the LFS, they argue are considerably higher because respondents can unknowingly report themselves in the public sector when really they are in the private sector according to National Account definitions. An example is employees of public sector bodies such as universities who incorrectly classify themselves as being in the public sector. Universities are, in fact, part of the private sector in the National Accounts as they are not controlled by government. Employees working for agencies and or contractors can also classify themselves as working in the public sector in the LFS when, in reality, because their employer is a private sector organization they should be allocated to the private sector according to the National Accounts definitions. (Hicks et al, 2005, pp.9-10) The scale of the discrepancy between the two data sources is obviously of concern because the public/private sector union density estimates we presented above and the primary data source we use below is the LFS. In what follows I classify the universities as part of the public sector. The issue of employees of private sector agencies working as subcontractors for public sector organizations and misreporting that they work in the public sector and other related issues will be examined below. 3. Benefits of belonging to a union higher wages. 5 Downloadable at 6

8 The declines in unionization rates that have occurred in both the US and the UK do not appear to have been accompanied by a large fall in compensation of union workers relative to non-union workers (Blanchflower, 1999; Blanchflower and Bryson, 2003, 2004). In Blanchflower and Bryson (2003) we drew three conclusions derived from the same data files used here but only up to First, the union differential in the US we found was higher on average than that found in the UK (18 per cent compared with 10 per cent). Second, the union wage premium in both countries was untrended in the years up to the mid-1990s. Third, in both countries the wage premium fell in the boom years since 1994/95. We speculated that it was too early to tell whether the onset of a downturn in 2002 will cause the differential to rise again or whether there is a trend change in the impact of unions. It is our view that most likely what has happened is that the tightening of the labor market has resulted in a temporary decline in the size of the union wage premium. Time will tell whether the current loosening of the labor market, that is occurring in both countries, will return the union wage premium to its long run values of 10 per cent in the case of the UK and 18 per cent in the case of the US. (Blanchflower and Bryson, 2003, p. 231). Table 4 reports estimates of the union wage gap for the UK, updates the estimates in my earlier papers and for the first time extends them to the public and private sectors. Here data from the LFS surveys are pooled, which generates a sample size of just over 710,000 individuals. Public sector and union status are both based on worker selfreports. I report the size of the union effect obtained in a log hourly earnings wage equation: the numbers reported in the table are the antilogarithms of the coefficients minus one to give the percentage effects because the dependent variable is in logarithms. The wage equations include a standard set of controls for each country age and its square, gender, race dummies, schooling variables and industry, year and region/state dummies. Estimates are reported separately for all of the years for the UK and for the first and second halves of the periods, for the public and private sectors, for men and women as well as for the five public sector groupings. Declining union density in the private sector is associated with a declining union wage gap, suggesting a diminution of union power in that sector. Declining density in the public sector is associated with a rising, or at least constant, wage differential. It is also true that high density in the public sector is associated with a higher union wage gap and low density in the private sector is associated with a smaller differential. Public sector unions are being more effective on the wage front than are private sector unions. Part B of Table 4 reports estimates from the 1983 General Household Survey (GHS). First it shows that in 1983 there was a significant union wage gap for the early 1980s, consistent with evidence presented in Blanchflower (1984, 1986 and 1991). In the public sector in 1983 the union wage gap was significant and positive but below that existing in the private sector, which contrasts with the evidence from the most recent data available in part A of the Table using the LFS for , which shows that the public sector 7

9 union wage gap as more than double that existing in the private sector. In 1983 female union wage gaps exceeded those for males in both the public and private sectors as is found in the most recent LFS data in part A of the Table. Table 4 suggests that there has been a decline in the union wage gap since 2000 with the more recent data we have available. The union wage gap is higher for women than for men; over the last decade the gap averages 17% for women and 8% for men. The public sector wage gap averaged 16% from compared with 11% in the private sector. The public sector gap is higher than that in the private sector for both men (9% and 8% respectively) and women (18% and 13% respectively). The finding of a union wage differential that is higher in the public sector than in the private sector is, at first glance, hard to explain in the UK where there are national wage agreements and Public Sector Pay Review bodies for most public sector workers. 6 One possibility is that higher quality workers are in unions and lower quality workers are not in unions and including education controls fails to pick up these differences; that is to say that the results are being driven by omitted variable bias. This implies there would be a positive selection effect into union membership. There is some supporting evidence that the quality of public sector workers has fallen over time when public sector wages fell relative to private sector wages. In an interesting recent paper Nickell and Quintini (2002) have shown, using age 10 or 11 test score percentile positions, that men who entered teaching or public sector general administration in the early 1990s had a significantly lower test score percentile rank than those who entered in the late 1970s. No such falls were found for police where relative pay did not fall. In the case of women there were no significant changes in percentile rank. Nickell and Quintini did not address the issue of changes in the quality of public sector workers by area or union status. Budd and Na (2000) have suggested a number of additional possibilities why a wage premium for members relative to covered members might exist. First, workers who unionize are more motivated or more prepared to stay with the firm and invest in firmspecific human capital. Second, nonmembers may be discriminated against in order to discourage free-riding. Third, probationary periods during which workers are not required to join unions may explain why nonmembers have shorter tenure and lower wages. Similarly, permanent workers may face pressures from union shop stewards to join unions while younger, temporary or probationary workers will not be targeted in the same way. Fourth, union members may have higher levels of human capital than nonmembers as a result of apprenticeship programs and other training. Fifth, covered nonmembers may be in weaker unions than covered members. Finally, union coverage might be measured with greater error than union membership. Budd and Na found evidence of a union membership wage premium among full-time private sector employees covered by union contracts in the US using OLS of between 12 and 14%. Allowing union membership to be endogenous yielded even higher estimates. They went 6 Most of the Pay Review Bodies undertake collective bargaining; a major exception is the Prison Service. 8

10 on to show that their findings could not be explained by differences in job tenure, unobservable characteristics and measurement error. I find similar results for the UK in both the public and private sectors. The union wage gap in the UK remains once controls are included for job tenure, age, permanent or temporary jobs, human capital and errors in measuring union coverage. Another possibility is that jobs that have been contracted out under Private Finance Initiatives (PFI), especially in health and education will be less likely to be unionized, and workers in these jobs will likely report that they work in the public sector whereas they are really in the private sector, as suggested by Hicks et al (2005). Any differential estimated here though would be a true mark-up as the public sector unions are successfully protecting their members' interests. 7 We should keep in mind, however, that these various public sector initiatives are of relatively recent origin and we saw from part B of Table 4 that there were substantial public sector differentials existing in 1983 when union membership, recognition and coverage in the public sector was much higher than exist today. 8 In Table 5 I examine the characteristics of public sector workers who are not members of unions with public sector defined based on their self-reports - using the latest five years of union data available from the Autumn Labour Force Surveys, Sample size is 282,000 and the sample weights are imposed. Sixty four percent of all public sector employees in the LFS are female; one in three female employees work in the public sector compared with 16% of men. Public sector organizations appear to make fairly extensive use of non-union labor in short tenured and frequently temporary jobs and these temporary workers have lower unionization rates than permanent workers. In 2003 permanent workers had a unionization rate of 30.1% compared with 18.1% for temporary workers. In the private sector the rates were 18.7% and 9.4% respectively while in the public sector 61.5% and 33.4% respectively. 9 The non-union public sector workforce, according to these LFS data, have shorter tenure, are less likely to hold a permanent job, works less hours per week, does not work overtime, are more likely to work at a small workplace and is less qualified on average than unionized public sector workers. Sixtythree percent of non-union public sector workers in the data had a permanent job that they had held for at least two years, compared to 85% of union workers. What occupations do these public sector non-union workers have? This issue can be examined using the LFS data, as there is a consistent set of occupation variables in each year. Data are weighted once again. 7 I am grateful to David Metcalf for suggesting these points. 8 One possibility is that the 1983 public sector result is being driven by the existence of a closed shop effect which still existed at that time. 9 Source: Trade union membership, 2003, July 2004 published by Department of Trade and Industry, UK, Table 8. 9

11 Public Private Non-union Union Non-union Union Managers and senior officials 8% 7% 18% 11% Professional occupations 16% 29% 8% 10% Associate professional/technical 17% 26% 11% 12% Administrative and secretarial 24% 15% 12% 9% Personal service occupations 17% 12% 6% 4% Elementary occupations 14% 7% 13% 11% Other occupations 6% 5% 32% 43% Private sector union members are generally in the less skilled occupations. More than half of union workers in the public sector are concentrated in Professional and Associate Professional occupations compared with less than a quarter in the union private sector. The two most important occupations in the union sector are teaching professionals (21.3% of public sector union workers) and health associates (11.0%). These two occupations account for a considerably smaller proportion of the non-union public sector workforce (7.8% and 2.6% respectively). There are differences in qualifications by sector in the private sector 27% of workers in both the union and non-union sectors had qualifications above A-level or ONC/OND. In the public sector 58% of union workers and 39% of non-union workers respectively had these qualifications. Public sector nonunion workers are spread across many occupations. In Table 6 I examine the characteristics of public sector workplaces using data from the Workplace Industrial Relations Survey (WIRS) Time Series Dataset, Data are weighted by the establishment weights and are restricted to workplaces that employed at least 25 workers (full plus part-time). This is a useful comparison because public sector status is determined at the level of the establishment and mapped in from the management questionnaire and not based on responses by the individual. To the extent that non-union public sector workers are employed in workplaces with under 25 employees, as suggested by Table 5, then any estimates derived from WIRS will likely underestimate any union/non-union wage differences. The WIRS data makes clear that the vast proportion of workplaces in the public sector continue to recognize unions in % of all public sector workplaces recognized unions, which is the same percentage reported in Coverage in 1998 was lower at 63% compared with 68% in Table 6 makes clear the extensive use of both workers on fixed term contracts in the public sector alongside the widespread use of subcontracting. More than half of all public sector workplaces made use of fixed contract workers in The growth has been most pronounced during the 1990s. 11 There has also been dramatic growth in subcontracting in the public sector workplaces. In 1990, 57% of public sector 10 Recognition was 99% in 1984 and coverage was 93% in the same year. 11 Data are also available for earlier years private sector 1980 and 1984=11%; public sector 1980=37% and 1984=36%. 10

12 workplaces used subcontracting but by 1998 that had increased to 82%, broadly comparable to the rate prevailing in the private sector. More than half of public sector workplaces subcontract cleaning and building maintenance while just under half subcontract catering. The second part of the table shows the growing use of subcontracting in the union public sector whether union is defined by recognition, high membership (>=75%) or high coverage (>=75%). Due to the possibility that the public sector union wage gap occurs because some individuals in occupations such as cleaners and catering staff who are working in the public sector but are actually employed as subcontractors in part A of the Table 7 using data from the LFS I control for occupation. These years were chosen as they are the most recently available and there is a consistent set of occupation codes across all four years. In Blanchflower and Bryson (2003) we argued that there was a potential problem with doing this as it was equivalent to regressing wages on percentiles of the wage distribution the highest occupation group is actually the 99 th percentile and the lowest the 1 st percentile. Barry Hirsch has taken a different view and argued that occupation dummies should be included because they reflect large differences in skill not controlled for by years of schooling and age. Given that there is a large variation in individual union status within broad occupation groups, his view is that they may be an appropriate control. In this case we are including occupation controls for a rather different reason, because of the possibility that the measured wage gap in the public sector isn t real and is simply due to measurement error in the public sector status variable. When occupational controls at the 2-digit level are introduced into the public sector equation they do have an impact on the results but there are still significant wage differentials. Introducing the occupation dummies lowers the size of the union wage gaps differentials from 15.1% to 9.4%. Even with the introduction of the occupation controls the estimated wage gaps in the public sector remain larger than those in the private sector. The next eleven rows of Table 7 explores what happens to the differential as first additional controls and secondly when the sample is restricted to a variety of subsamples. Line 4 now adds a foreign born and a fulltime dummy while line 5 adds 5 workplace size dummies and 5 temporary work type dummies (seasonal, casual, fixed term contracts, temporary agency and other temporary). Even though the addition of these controls lowers the size of the estimated differentials they are still substantial and significant. The addition of a deeply endogenous job tenure variable does not drive away the significance of the union variable; it is endogenous as unions can control entry to the job. Line 8 restricts the sample to temporary workers only and line 9 to permanent workers and the results for the two groups are very similar. Restricting the sample by years of job tenure produces significant differentials until 25 years of tenure and over. There is some evidence that the differential declines as tenure rises. Lines suggest that the differential declines with size of workplace. There is a great deal of evidence to suggest that what we are observing is real and is not simply limited to low level occupations misreporting public sector status. 11

13 Further evidence for the view that the public sector differentials are real and not caused by measurement error or omitted controls for job tenure, apprenticeships, human capital etc. is provided in part B of the Table where separate results are presented for the major one digit occupations in both the private and public sectors. I do not report results for sales and customer service and process plant and machinery operatives in the public sector as sample sizes are small, for obvious reasons. For each of these occupations there is a significant union wage gap for all occupations in the public sector and for all but the professional occupations in the private sector. The differentials are highest in the private sector for the least skilled occupations: there is less variation by skill level in the public sector. There are substantial wage differentials for the highest level managerial and professional occupations, controlling for their education, gender, race, location, industry and age. The fact that there are significant differentials for each public sector occupation suggests that there is something real going on. The significant public sector union wage gaps estimated here do not appear to be an artifact of the data. I re-estimated the equation in row 6 of Table 5 to separate out private and public sector union and non-union workers and in the first column below included only three dummy variables to represent these categories plus year dummies. In the second column below the following controls were included - a foreign born dummy, a fulltime dummy, 5 workplace size dummies, 5 temporary work dummies, age and age squared, male, 4 race dummies, 40 schooling dummies, 19 regional dummies, 60 industry dummies and 3 year dummies. The final column adds the complete set of 81 occupation dummies. The results were as follows the excluded category is public sector union with t-statistics in parentheses (n=61,093). Percentage effects are reported after taking antilogs. Only year Full set of controls Plus occupation dummies dummies Private union 17.7% (18.33) 11.6% (15.28) 12.2% (16.55) Private non-union -2.3% (18.33) 5.8% (9.61) 3.0% (5.14) Public union 32.6% (33.37) 12.1% (16.87) 11.2% (16.86) It is clear that once individual and workplace controls are included public sector nonunion workers are the lowest paid. Once personal and workplace controls are included there is no significant difference between the earnings of private union workers and public union workers (t=1.25 on the difference in column 3). The bad jobs are held by public sector workers who are not members of unions. It is still possible that the finding that public sector non-union workers are the lowest paid arises because of unmeasured quality differences both of workplaces and individuals. One possible way to get at the issue is to control for workplace characteristics and this can be done using weekly wage data from the 1998 WERS. I estimated union/non-union wage differentials at the level of the individual worker in both the public and private sectors. It is unlikely that public sector status is miss-measured in these data given that ownership status is derived directly from reports of the manager interviewed at the workplace where the worker was employed which is then mapped onto the individual 12

14 worker data file. 12 Workers on subcontracts employed by private sector firms will be excluded from this analysis because the wage data specifically relate to employees at that workplace. In part A below control variables included 5 schooling dummies, hours of work, workplace size, 11 industry dummies, gender, 6 age dummies and 6 race dummies. In part B a full set of 1780 establishment controls are included and, hence, establishment size, industry dummies and private sector dummy are excluded and the method used is random effects interval estimation. Identification of the union wage gap then comes from within workplace variation. Union status is determined based on the reports of the individual worker. Because the wage data are in bands and open ended, interval regression is used, with the higher open-end closed off at 750 as in Booth and Bryan (2004) who also used these data in an analysis of the private union sector. Separate results are presented according to whether the individual says they are in a permanent job or a temporary job or one on a fixed-term contract. The results were as follows (n=26,104; n=17,022 in the private sector and n=9,082 in the public sector). All Private sector Public sector A) OLS All 8.2% 5.0% 12.9% Permanent 8.3% 4.7% 12.5% Temporary/fixed term 17.4% 15.5% 17.3% B) Random effects All 12.9% 11.7% 16.2% Permanent 12.1% 11.9% 15.8% Temporary/fixed-term 19.1% 20.2% 17.2% Using the random effects procedure raises the size of all estimated differentials except for that of temporary workers in the public sector which remains largely unchanged and large. Public sector union wage gaps remain significant, substantial and higher than those in the private sector. The question we now turn to is whether these public sector union wage gaps occur at covered workplaces. Booth and Bryan (2004) examined whether there were union wage effects within covered private sector British workplaces using these same data and argued that they could find none, once they included workplace specific fixed effects. 13 The implication from their work if there is no member premium is that a free-rider problem exists nonmembers are 12 The primary unit of observation in the WERS is the establishment, with a minimum cut-off of 10 employees. Survey questionnaires were also sent to 25 randomly selected employees at each workplace or all employees in smaller workplaces. The questionnaire was distributed to the 1887 workplaces where management permitted it, with a response rate of 64% (28,237 employees). 13 Hildreth (2000) found similar results for the private sector using worker-reported coverage data from waves 1-4 of the British Household Panel Study Worker reports of coverage are inherently less reliable than reports taken from employers. 13

15 obtaining the same benefits as members without paying the dues. I use their procedure with one or two modifications and a larger sample size and examine the results in both the public and private sectors using the same data source and can find absolutely no support at all for their contention. 14 The union coverage variable Booth and Bryan used was derived as follows. For each of nine occupations management was asked "which of the following statements most closely characterizes the way that pay is set [for each occupational group present at the workplace]?" Options given were as follows. Collective bargaining for more than one employer (for example, industry-wide agreement). Collective bargaining at an organization level. Collective bargaining at this workplace. Set by management at a higher level in this organization. Set by management at this workplace. Negotiation with individual employees. Some other way (for example, pay review body). None of these. Booth and Bryan defined a group as covered if a manager selected any of the first three statements to describe an occupational group; we use these criteria in the private sector but in the public sector we also include responses from some other way because of the existence of pay review bodies in the public sector for large numbers of workers. The question addressed to workers in the Survey of Employees asks, which of the following occupation groups best describes your job at present? and is followed by a list (with examples) of the nine one digit occupations. These individual replies were matched to the workplace occupation level coverage status variables derived from the management responses, to obtain a coverage indicator for each worker. According to these measures 26% of private sector workers and 87% of public sector workers in the WERS data set were employed in covered workplaces. Booth and Bryan used interval regression with fixed effects but were concerned that with the non-linear methods required for interval regressions the coefficient estimates would be inconsistent although they did note that in practice the bias is likely to be small. As an 14 There are several other differences between the procedure used here and that used by Booth and Bryan (BB). a) BB used log hourly earnings as the dependent variable. I used log weekly earnings but included hours of work as an RHS variable. b) BB included a number of endogenous controls including presence of health problems, marital status, dependent children and job tenure that I exclude. c) BB restrict their sample to full-timers and exclude workplaces of less than 25 workers and I do not. d) BB report results for four groups of workers male manuals, female manuals, male non-manuals and female non-manuals. They do not report any results for the four groups combined. This results in small sample sizes and very few degrees of freedom especially when approximately 200 workplace fixed effects are included. Sample sizes for these four groups in their Table 5 are 1,357; 441; 805 and 783 respectively. 14

16 alternative they used a procedure that involved adding as controls the means of the individual variables in the workplace; both methods produced very similar results. I use the random effects interval regression model because there is not a sufficient statistic that allows the fixed effects to be conditioned out of the likelihood. I also present results using OLS fixed effect estimation. The samples below are restricted to covered workers (n=4,114 in the private sector and n=6,598 in the public). Controls are 6 age dummies, gender, 4 race dummies, 8 occupation dummies, 3 temporary worker dummies, 5 schooling dummies, an apprenticeship dummy, hours of work and 11 industry dummies. The dependent variable is the log of weekly earnings. In contrast to Booth and Bryan I did not restrict my sample to larger establishments or exclude men with low hours. I also excluded the various endogenous regressors Booth and Bryan included job tenure, health problems, marital status and the presence of dependent children. All union coefficients below are highly statistically significant and are the percentage effects after taking antilogarithms. T- statistics are in parentheses. Results are reported using interval regression methods with random effects and OLS with fixed effects. Private Public Random effects All 10.2% (4.10) 12.6% (5.77) Permanent 10.0% (3.94) 12.7% (6.05) Temporary/fixed-term 10.0% (1.75) 8.9% (2.38) Full-time male manuals 3.9% (2.15) 8.3% (3.58) Female manuals 16.8% (3.49) 8.6% (3.53) Full-time male non-manuals 4.7% (2.73) 4.6% (3.25) Female non-manuals 6.2% (1.20) 5.6% (2.85) Fixed effects All 5.9% (4.54) 16.1% (13.62) Permanent 5.3% (3.98) 16.5% (13.67) Temporary/fixed-term 11.4% (1.54) 1.9% (0.34) Full-time male manuals 3.4% (1.81) 7.4% (2.67) Female manuals -3.0% (0.43) 14.8% (3.31) Full-time male non-manuals 2.0% (0.90) 5.5% (3.00) Female non-manuals 5.2% (2.28) 13.5% (8.53) I find evidence in both the private and public covered sectors that, apart from female nonmanuals in the private sector, members earn significantly more than nonmembers. Booth and Bryan find that there are no such effects in the private sector with these same data; I am unable to replicate their findings or sample sizes even when I separate out the sample to the four subgroups they use and drop establishments of under 25 employees as they 15

17 do. 15 The interpretation here is clear though: there are significant union wage differences within both private and public sector unionized workplaces, but unexpectedly they are generally larger in the union public sector than in the unionized private sector for permanent workers. Union wage differences of around 13% are present in covered public sector workplaces where the coverage agreement in question relates to the non-union worker s exact one-digit occupation. These differences remain once detailed controls are included for industry, race, gender, fulltime status, whether the job was temporary or permanent occupation, job tenure, human capital, age, workplace size, training etc. The findings of a union wage gap in the public and private sectors as well as the finding that the differential is higher in the public sector is robust to the myriad of specification checks I threw at the data. The exact reason for the differential remains unclear it may well have to do with unobserved quality differences of workers perhaps in terms of motivation or other unmeasured productivity enhancing characteristics - or discriminatory actions by unions against nonmembers as suggested by Budd and Na (2000). Unfortunately in these data discrimination cannot be addressed directly. There are also no obvious instruments available that are correlated with the endogenous union membership variable yet uncorrelated with the wage rate that would allow me to model possible endogeneity or selection bias into union membership. Such work that has been done on the issue suggests that the OLS results cannot be explained by nonrandom selection. There is evidence that accounting for union membership endogeneity yields union wage gap estimates that are higher than OLS estimates, implying negative selection into union membership where those with lower ability choose to be members to raise their wage (see Budd and Na, 2000 and Hildreth, 2000). Analysis of longitudinal data at the individual level allow for unobservable fixed effects but there are difficulties with measurement error that can bias such estimates downwards (Freeman, 1984 and Hildreth, 1999). It does appear that unions are able to raise wages by significant amounts in the order of 10-15% - in both the public and private sectors. It is somewhat surprising to find such high union differentials within the public sector given the high levels of union recognition and coverage existing across government workplaces. It appears that there are groups of non-union workers, some of whom are on temporary contracts who are outside these union agreements. Some of these non-union workers in the LFS say they work in the public sector but are apparently employed by private sector firms. The results remain once we overcome the misreporting problem. What is clear though from the evidence from the 1998 WERS is that there is a remaining group of non-union workers in the public sector who are paid significantly less than unionized public sector workers who are employed in the same occupations at that workplace. It remains a puzzle why these nonmembers do not join unions given the potential benefits open to them as they are currently not receiving the covered wage. It is possible, of course, that they are receiving other non-wage benefits that we have been unable to measure. There is little or no evidence of a free-rider problem in the UK covered sector. As Budd and Na (2000) 15 Booth and Bryan (2004) restrict their sample to full-time males and full-time and part-time females and excluded workplaces of less than 25 workers. 16